AI continues creeping into the clinic as it diagnoses skin cancer better than expert dermatologists, new research shows. What’s next?

geralt / Pixabay

The News:

Skin cancer was found to be diagnosed more accurately by artificial intelligence than experienced dermatologists in a new international study. Researchers tested a form of machine learning known as a deep learning convolutional neural network (CNN) to reach this conclusion. The study, titled “Artificial intelligence for melanoma diagnosis: How can we deliver on the promise?” was published in the cancer journal Annals of Oncology on May 28, 2018.

Malignant melanoma accounts for 1% of all skin cancers but causes a majority of skin cancer-related deaths Medical Daily reports. The American Cancer Society estimates 9,320 people will die from melanoma in 2018 while 91,270 new cases will be diagnosed. As part of the new study, researchers from the US, Germany, and France decided to test the performance of a CNN in diagnosing malignant melanomas by looking at images of moles.

“The CNN works like the brain of a child,” explained study author professor Holger Haenssle from the University of Heidelberg, Germany.

More than 100,000 images of malignant and benign skin cancers and moles were shown to the network along with the diagnosis for each image.

“Only dermoscopic images were used, that is, lesions that were imaged at a 10-fold magnification. With each training image, the CNN improved its ability to differentiate between benign and malignant lesions,” Haenssle added.

Once the network was trained, two sets of images were built using new pictures never seen by the CNN. The first set of 300 images was to test the abilities of the CNN alone, while the second set of 100 images was to test both the network and a group of doctors.

Fifty-eight dermatologists, from 17 countries, agreed to take part in the test of man versus machine. The number of skin cancer cases missed by the CNN were fewer than those gone unnoticed by dermatologists, indicating a higher sensitivity. The misdiagnosis of benign moles as melanoma also saw a lower rate with the network, which could help avoid unnecessary surgery.

Looking at the numbers, on average, the dermatologists correctly detected around 86.6% of melanomas while the CNN identified 95% of them. After the dermatologists were provided clinical information about the patients such as their age, their sex, and the location of the lesion, their success rate in diagnosing melanoma increased to 88.9%.

“When dermatologists received more clinical information and images at level II, their diagnostic performance improved,” Haenssle said. “However, the CNN, which was still working solely from the dermoscopic images with no additional clinical information, continued to out-perform the physicians’ diagnostic abilities.”

The researchers do not believe the CNN will replace human professionals, but rather, be used as a tool to reduce the risk of misdiagnosis. They added most dermatologists were already using digital dermoscopy systems and other such tools for cancer-related documentation, detection, and follow-up of patients.

Steve’s Take:

I’ve been a long-distance runner for going on 45 years. It all began when I moved to Los Angeles from Washington DC and fell in with a group of hard-core marathoners. There were from around 10 of us on any given Saturday morning when we would meet on the tree-lined median strip of San Vicente Blvd. and Fourth Street in Santa Monica.

Doing our version of the movie, “Chariots of Fire,” we would amble through Rustic Canyon, then Brentwood, cross Sunset Boulevard and then climb up to Will Rogers State Park, finally reaching the summit–Inspiration Point. We’d circle that little area until everyone caught up and then descend back down to San Vicente–a total of maybe 11-12 miles.

We were mostly in our 30s and 40s and many of us ran shirtless, some even capless. If we only knew then what we know now. Every last one of us comprehends the perils of all of that radiation, which although unquestionably drop-dead gorgeous, would eventually introduce us to the various forms of skin cancer later in our lives, including the dreaded and unfriendliest, melanoma.

Having said all that, most of us see our dermatologist at least every year, if not every six months (in my case) to have various lesions frozen, burned off, or surgically excised.

Now we are being told that that an AI called CNN is better than dermatologists at detecting malignant melanomas. No, not the cable news network, but convolutional neural network (CNN), which is a type of AI.

In this study, the majority of participating dermatologists weren’t new to the industry, either: just 17% had less than two years’ experience in dermatoscopy, while 19% had two to five years and 52% of the dermatologists had more than five years’ experience.

Importantly, the dermatologists were able to detect more melanomas when they had context about the patients’ background–88.9% of malignant melanomas and 75% of non-malignant lesions–yet, this still wasn’t as accurate as the CNN.

“These findings show that deep learning convolutional neural networks are capable of out-performing dermatologists, including extensively trained experts, in the task of detecting melanomas,” Holger Haenssle of the University of Heidelberg and first author of the study, explained.

AI has been making huge inroads into England’s National Health Service, and in the detection of cancer in recent years. Last year, Stanford University created an AI that spots the warning signs of skin cancer with 91% accuracy. Then in October, an AI created in Japan was able to identify cancer just through a colonoscopy.

Bottom Line:

Although the winner in this latest scientific contest was CNN, Bruce Lee for Forbes points out several important caveats. It’s going to be a while before we can expect to have a robot scanning our bodies with the dermatoscope and no longer encounter a human dermatologist.

Remember also that there is substantial variability in the skills and abilities of dermatologists, or any doctor for that matter. For example, the standard deviation for properly identifying a melanoma was over 9% for the dermatologists and for properly identifying a benign mole was over 11%. Thus, it’s not absolutely clear if CNN is currently better than some of the best dermatologists around.

The 58 dermatologists were not necessarily the “best of the best” but simply those who responded to an invitation sent to the International Dermoscopy Society mailing list. So, the only thing clear was that they were comprised the group that responded to this particular email invitation.

Also, showing a dermatologist an image is not quite the same as a dermatologist seeing and talking to the person attached to the lesion, Lee notes. Lesions may be more or less difficult to evaluate depending on where they are located on a person’s body. And, of course, people’s skin varies by skin color, texture, hairiness, and many other characteristics.

I concur with those, like Lee, that this finding doesn’t mean that AI should ever replace dermatologists completely. In other words, don’t expect a website or app to replace a doctor’s visit anytime soon. Nothing can beat having an experienced doctor carefully looking you over, talking to you, getting your clinical history, and making an informed decision about what to do next.

Steve's Take: #AI could lead to a far more efficient and less error prone #healthcare delivery system that is a lot less burdensome and costly Click To Tweet

In general, AI and other computational approaches have the potential of relieving doctors of repetitive tasks that can be done more efficiently or precisely by computers and allowing doctors to focus more on work that require uniquely human abilities and qualities. This includes spending more time interacting with patients, managing health care teams and operations, and developing and guiding ways to improve patient care.

Final Thoughts:

In reality, the biggest difference between physicians is not their level of intelligence, but (a) how they approach patient problems and (b) the health systems that support them. And because “a” and “b””combine to create wide variations in clinical outcomes nationwide, machine learning offers great hope for the future, says Robert Pearl, MD, for Forbes.

There are many who fear machines will replace (or even turn on) humans. I’m in the camp of those who believe these fears are grounded more in science-fiction than reality. It’s true that computer intelligence is advancing faster than human intelligence. But this development offers far more opportunities than perils.

If we see computer speeds double another five times over the next 10 years, machine-learning tools and inexpensive diagnostic software could soon become as essential to physicians as the stethoscope was in the past. And financing a far more efficient and less error prone healthcare delivery system could get a lot less burdensome and costly. I foresee a lot of winners in this process of AI creep. Sorry Arnold, you’re not in this particular picture.

Print Friendly, PDF & Email